Abstract
<title>Abstract</title> <p>Hybrid RF/FSO/THz communication systems provide solutions with different reliability levels and capacity options and energy efficiency performance, the system performance requires adaptive channel selection because environmental factors, including rain, fog and atmospheric turbulence. This paper presents a channel-aware framework that enables hybrid RF/FSO/THz systems to adapt their communication based on physics-based modeling and data-driven decision processes. The study creates a multi-scenario synthetic dataset which uses realistic propagation models to simulate environmental changes and random channel behaviors. The authors use supervised learning models and reinforcement learning agents to determine the best communication channel for different operational conditions. The proposed approach undergoes evaluation through conventional evaluation methods and cross-scenario generalization testing, which involves training models on specific environmental conditions and testing their performance on new scenarios. The study demonstrates that high accuracy exists under matched conditions, but performance decreases when system distribution changes, which demonstrates the need for diverse scenarios in training. The additional experiments which used noisy channel observations proved that the models exhibit strong reactions to changes in measurement error. The learning-based models demonstrate their ability to simulate a physics-based channel selection policy through their performance in simulated environments which create a controlled testing foundation for upcoming real-world tests. The findings reveal new information about the limitations of learning-based channel selection systems in hybrid wireless networks while showing their practical challenges, which need strong decisionmaking methods that work across different scenarios.</p>